Empirics and the Pollution Haven Hypothesis (PHH)

1. Empirics and the Pollution Haven Hypothesis (PHH) November 10, 2007

2. Empirical questions related to PHH • Do investment flows respond to differences in environmental standards? • Has trade liberalization increased pollution intensity in developing countries? • Have tighter standards in developed countries led to loss in pollution-intensive industries? • The literature does not attempt to determine whether countries use environmental policies that are too weak, in order to attract investment or increase market share of dirty goods. That is, the literature does not attempt to uncover the motive of environmental policy.

3. What is a statistical model? • We are interested in relation between net exports and pollution control costs. • We know that net exports depend on other variables (e.g. supply of factors – remember the HOS model and Rybczynski theorem) • If we have data on these variables we can estimate a relation between exports and pollution control costs, while “controlling” for other variables (e.g. supply of factors). • We are (usually) interested in sign and magnitude of coefficient on pollution control costs, and on whether the coefficient is statistically significant.

4. More details on statistical model (a.k.a. “regression equation”) • The subscript i identifies the country and the subscript t identifies the time period. For the PPH, the dependent variable y is a measure of exports of the dirty good, the explanatory variable x is a measure of pollution control costs, z contains other explanatory variables, called “control variables” (e.g. factor endowments for the PHH); e is the equation “error”, a composite of factors that we do not observe, but which affect the dependent variable, and a component that takes into account the inherent randomness of the process. • The statistical problem is to estimate the parameters, particularly beta, and determine whether it is positive and statistically different than 0. • There are many technical problems: missing data, data with measurement errors, correlation between error and explanatory variables, misspecification of model….

5. Alternatives for addressing question “Does trade harm the environment?” • Theory, i.e. try to determine the likely relation between trade and the environment using logic. Theory helps you “think clearly” but is inconclusive. • Case studies, i.e. finding examples where the relation appears positive or negative. These are useful, but they leave you wondering how representative the case studies are. • Statistical models – these have the advantage of being based on widely accepted principles, but the data seldom exactly conforms to the statistical assumptions.

6. The empirical evidence • Early studies use US data to categorize industries into dirty and clean sectors (based on emissions per $ of output, or per employee, or on abatement costs). • The statistical exercise looks for link between dirty and clean good trends in production or export (share) and country characteristics such as income, income growth, and openness. • Are developing countries “moving toward” dirty industries? • This type of exercise ignores possible changes in technique -- it assumes that changes in composition translate directly into changes in pollution. Also ignores other explanatory values, such as factor endowments.

7. Early evidence • Early research found that a rise in environmental control costs in North was positively correlated with increases in dirty good share of exports from developing countries, and decreases in dirty good share of exports from rich countries. • The Lucas and Wheeler study found that toxic releases per unit of output (measured by GNP) fell as countries became richer, due to changes in composition. Poorer countries had the largest increases in toxic intensity. • Birdsall and Wheeler found that pollution intensity increased most rapidly in Latin American countries after OECD pollution regulation became stricter.

8. Interpretation of these results • These findings are consistent with PHH, but are also consistent with an explanation based on changes in factor endowments (capital accumulation). • Evidence for the importance of capital accumulation: (i) Over 90% of dirty good production in 1988 was in OECD countries, suggesting that location of dirty good production reflects more than weak environmental regulation. (ii) If stricter environmental policies in rich countries were responsible for reallocation of dirty good production (as in PHH) then we would see an increase in the relative price of dirty goods; if capital accumulation in South caused the reallocation, the relative price would fall. Data does not show a clear upward or downward trend in relative price. (iii) All studies show that poor countries alter their mix of production toward dirty goods, the more open countries have a cleaner mix. Pollution intensity grew most rapidly in the more closed economies.

9. Early studies of trade effect of pollution control costs • Tobey uses cross country data on exports of 5 dirty commodities and country-specific factor endowments and measures of environmental stringency. • Few “degrees of freedom” (not much data). Coefficient on environmental stringency insignificant, but so are most of the coefficients on factor endowments. • The statistical model does not explain much …of anything.

10. The relation between trade flows and measures of environmental stringency • Link net exports (as share of value of industry production) to industry-specific measure of environmental stringency (e.g., abatement costs) and industry characteristics (such as cost shares of labor, capital, and maybe tariff rates). • The PHH implies that the coefficient on the environmental stringency variable should be negative (more stringent environmental policies lower net exports.) • Studies do not find a significant negative relation between environmental stringency and net exports.

11. Statistical reasons why these studies might incorrectly reject PHH • Small sample leads to lack of statistical significance. • Several reasons why models might produce biased estimates: • Measurement error • Endogenous explanatory variables • Omitted explanatory variables that are correlated with included variables • Three examples follow. In the first, a statistical model correctly identifies relation between pollution control costs and trade. In the second two examples, statistical model leads to biased estimates. The bias could go in either direction.

12. What do we mean by speaking of the “demand function” and the “supply function” for pollution? • The demand function: Think of pollution as the use of the environment as a dumping ground. Firms “demand” pollution (i.e. they want to use the dumping ground more) because it is cheaper for them to dump than to clean up or prevent pollution. A higher pollution tax (the price of dumping) decreases firms’ demand, so this function slopes down. • Pollution creates a cost to society. Society’s “supply function” for pollution equals society’s marginal cost of pollution. If the marginal cost increases, society’s “supply function” for pollution slopes up.

13. The optimal pollution tax (the price of a unit of pollution) is given by the intersection of the supply and demand curves for pollution Pollution “price”, equal to the tax Society’s supply function for pollution Firms’ demand function for pollution Pollution quantity

14. Example 1: statistical model correctly identifies a relation between pollution control costs and trade • Draw a downward sloping (industry) demand curve for pollution (the pollution tax is on vertical axis). A lower tax means that firms’ “demand” for pollution increases. • Suppose that there is an exogenous increase in pollution tax (maybe preferences become more green). The higher tax increases abatement costs per unit of output. • Since production costs (inclusive of abatement) increase, domestic supply (as a function of output price) shifts in. • At a constant relative commodity price, net exports fall. Here more stringent policy lowers exports (or raises imports) of the dirty good, as the PHH predicts. (See next slide.)

15. A higher tax reduces firms’ level of pollution (left panel), increasing their production costs, causing the supply curve to shift in (dotted curve in right panel) tax Price of dirty good tax Firms’ “demand” for pollution Firms’ supply function for dirty good pollution Quantity of dirty good

16. Example 2: statistical model understates relation between abatement costs and trade (or gets sign wrong), due to an omitted explanatory variable that is incorporated into the error term – leading to correlation between the pollution tax and the error (a form of endogeneity) • Suppose that the pollution tax is endogenous; it is determined (optimally) by the intersection of a (industry) demand and (society’s) supply function for pollution. • An increase in a factor (e.g. capital) used intensively in polluting industry shifts out demand curve for pollution. This variable is not included in the statistical model, so it gets incorporated into the error term. • This change leads to a higher pollution tax, and higher abatement costs. • However, the increase in the factor also shifts out domestic supply function of dirty good. (Higher tax and larger supply of factor “cut in the opposite direction.) Net exports increase. • Here pollution taxes are positively correlated with net exports, contrary to PHH. • The higher pollution tax does not cause the increased export of the dirty good. Instead, the exogenous growth in a factor leads to higher output of the dirty good and to higher pollution costs. • The higher tax does reduce exports (since the relative supply curve would have shifted out more in the absence of the tax increase.) • This measurement problem would not arise if the statistical model included the missing variable (the stock of capital in this example).

17. Increase in factor shifts out demand curve for pollution (dotted curve, left panel), raising the pollution tax. By assumption, the higher supply of factor decreases marginal cost of dirty good, even with the higher tax, so supply function of dirty good shifts out (dotted curve in right panel). A higher pollution tax is correlated with higher supply of dirty good. Dashed curve right panel shows the supply effect of increase in factor, absent the increase in tax Price of dirty good tax Pollution quantity Dirty good Supply and demand of pollution Supply of dirty good

18. Example 3: statistical model overstates relation between abatement costs and trade, due to omitted explanatory variable • The statistical model regresses exports on pollution abatement costs, but omits transportation costs. PHH suggests that high abatement costs discourages domestic production in dirty sectors, so sectors with higher abatement costs would export less. • Dirty industries (in this example) have higher transport costs (e.g. cement) relative to clean industries. • High transport costs discourage exports. Suppose that transport costs (e.g. energy costs) increase. • The higher transport costs have a disproportional effect on dirty goods (because transport is more important in those sectors). The higher transport costs have a disproportional effect on exports of the dirty goods. • In this case we have an excluded variable (transport costs) that is positively correlated with an included variable (abatement costs). • In this example, the estimate on the coefficient of abatement costs is upwardly biased. Here the statistical results exaggerate the trade effect of abatement costs.

19. Some recent statistical evidence • Evidence from US studies shows that these endogeneity and missing variable issues might be part of the explanation for the failure of statistical evidence to support the PHE. Intra-US trade data is better than world trade data. • US studies estimate the relation between investment (into US states or counties) and measures of environmental stringency. When the studies account for endogeneity and heterogeneity, they often find a significant negative relation between inward investment and abatement costs, as the PHH suggests. • In other words, correcting for endogeneity and other statistical problems might uncover stronger evidence for PHH. • Even if stronger pollution control alter investment and trade flows on the margin, it is unlikely to be strong enough to offset other considerations, such as factor endowments. • PHH may be more important in future. Pollution abatement capital expenditures have risen from 2.8% of new capital expenditures in US in 1984 to 7% in 1993.

20. Related (older) trade and environment studies • Grossman and Krueger estimated that the “composition effect” of Mexico joining NAFTA would likely reduce pollution. • This composition effect appears to have actually occurred. However, it was swamped by “scale effect” (increased aggregate production), leading to increased pollution in Mexico. • Of course, we do not know that this higher pollution was a consequence of NAFTA.

21. Summary • Trade is determined by many things (e.g. factor endowments, technology, infrastructure, institutions). • Costs of environmental measures are small in most sectors, so they likely have only small effect on investment decisions and trade flows. • There is some (emerging) statistical evidence that identifies these small effects. • Environmental costs and cost differences might increase over time, (e.g. next version of Kyoto Protocol), making PHH more important in the future. • In some sectors these costs are already large enough to effect pattern of trade (e.g. battery disposal, ship breaking). Basel Convention can regulate trade in these sectors (better than general trade restrictions).

22. Summary, continued • There are many reasons why countries have different levels of environmental protection. (Differences in competing needs and constraints, preferences, assimilative capacity.) • Statistical evidence cannot determine the rationale for level of environmental protection. • In addition to (possibly) reallocating production of dirty goods from rich to poor countries, globalization is (plausibly) associated with income growth and technology transfers that at least offer the opportunity of environmental improvements.

23. And most importantly… • Trade policy is a poor substitute for environmental policy. • Remember the Principle of Targeting.